Sharedbackup_liu2013

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication.

Survivable Traffic Grooming in Elastic Optical


Networks - Shared Protection
Menglin Liu, Massimo Tornatore, and Biswanath Mukherjee

Abstract—This study investigates the survivable traffic groom- flow on a lightpath decreases, some frequency slots it occupied
ing problem for elastic optical networks with flexible spectrum before can now be released and used by other lightpaths [7].
grid employing new transmission technologies. In such networks, In optical networks, the failure of a network element (e.g.,
instead of following the traditional fixed ITU-T wavelength grid,
optical transponders are capable of properly tuning their rates, a fiber cut) can cause huge data loss, resulting in the failure
and consequently their spectrum occupation, by introducing the of several lightpaths. This problem becomes more compelling
fine-granular spectrum unit, called a frequency slot. The number when lightpaths are migrated to high bit rates, such as 40, 100
of contiguous frequency slots allocated to an optical path (i.e., Gbps and beyond, which are expected to be accommodated
lightpath) is adjusted to the current network flow. In this study, in future elastic networks. Hence, survivability in elastic
we propose a novel shared protection specific to elastic networks,
namely, elastic separate-protection-at-connection (ESPAC). It not networks is a critical issue. Shared protection has been studied
only provides traditional backup sharing, but also offers a new with static traffic (where a traffic demand matrix is known
opportunity of spectrum sharing enabled by the elasticity of the [8]), and for dynamic provisioning using online approaches
transponders: 1) if the working paths of two connections are [9]. Ref. [10] proposed dedicated protection in transparent
link disjoint physically, and 2) if their backup paths traverse two flexible-grid network, showing better spectrum utilization by
lightpaths which are adjacent on a fiber link, then the two backup
lightpaths can share spectrum. The new opportunity of spectrum flexible bandwidth assignment. Ref. [11] proposed a best-effort
sharing is realized by using First-Fit to assign working traffic restoration while satisfying the service level specifications,
and Last-Fit to assign backup traffic, and allowing spectrum taking advantage of the elastic bandwidth variation.
overlap between adjacent backup wavelengths. The elasticity of However, most studies on elastic networks are based on the
the transponder enables the expansion and contraction of the assumption that one lightpath is provided for each connection
lightpaths, thus when a single failure occurs in the network,
lightpaths carrying backup flows can be tuned to appropriate request [6]–[12], which cannot guarantee optimum spectrum
rates in such a way that the overlap spectrum is used by only utilization because a fair amount of spectrum is wasted as
one of the adjacent lightpaths. The results show ESPAC is very guard bands. To further improve the spectrum efficiency, we
spectrum efficient in elastic network setting. recast in elastic networks the classical problem of protected
Index Terms—Shared protection, Elastic optical network, Traf- traffic grooming in WDM networks [13], which allows us to
fic grooming, Dynamic provisioning, Spectrum assignment. multiplex different connections on one lightpath. Therefore,
grooming capabilities for transit signals must be incorporated
in the bandwidth-variable wavelength cross-connects (BV-
I. I NTRODUCTION
WXCs) to switch elastic lightpaths [1]. Note that the problem
The term “elasticity” refers to a set of recent innovations of traffic grooming in elastic networks is a nascent research
in optical networks. In an elastic optical network, rather field and has been so far considered only in unprotected
than following the ITU-T fixed grid with its rigid 50/100 networks, e.g., [14] proposed an electrical-traffic-grooming
GHz bandwidth, appropriate-sized optical bandwidth in sub- network design and showed that it saves at least 8% average
and super-wavelength granularity is allocated to an end-to- spectrum utilization compared to the non-traffic-grooming
end path [1], achieving significant gains in network design approach. The work in [15] proposed an optical-grooming
and operation in terms of CAPEX, and OPEX [2]–[4]. Such network design where they groom the traffics from the same
evolution is particularly beneficial for the increasing hetero- source onto one lightpath, and drop or switch optically a subset
geneity of today’s traffic, thanks to the new advances in of frequency slots at the intermediate nodes along the route
transmission technologies (e.g., new modulation techniques, to their corresponding destinations. Traffic grooming can also
such as optical OFDM, coherent detection, and advanced dig- lead to a saving of energy consumption, according to [3].
ital signal processing) that promise to provide flexibility [5], In this study, we propose to use First-Fit to assign spectrum
reconfigurability, and higher network agility. Such advances for the working paths, and Last-Fit to assign spectrum for
facilitate the flexible expansion and contraction of lightpaths the backup paths. We allow spectrum overlap between backup
(which are therefore referred to as elastic lightpaths) according wavelengths. Based on these spectrum assignment schemes,
to the traffic volume and user request [6]. Thus, when the we propose a protection approach – elastic separate-protection-
at-connection (ESPAC), providing end-to-end protection at
Menglin Liu (mlliu@ucdavis.edu) and Biswanath Mukherjee are with the connection level [16]. A path that carries traffic during normal
Department of Computer Science, University of California, Davis, CA USA. operation is called a working path. When any fiber link on the
Massimo Tornatore is with the Department of Computer Science, University
of California, Davis, CA USA, and with the Department of Electronics and working path fails, the connection is rerouted on a backup
Information, Politecnico di Milano, Italy. path. Under ESPAC, the spectrum of a lightpath can be used

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

Fig. 1. An example of FFW-LFB.

by either working paths or backup paths, but not both. If the


working paths of two connections are link disjoint, and their
backup paths traverse two lightpaths, which are adjacent on
a fiber, then the two backup lightpaths can share spectrum.
To the best of our knowledge, this study is the first to
investigate this type of backup-sharing approach and its impact
considering traffic grooming in elastic networks.
The rest of this study is organized as follows. In Section II,
we present the proposed ESPAC in elastic networks. Section
III presents heuristic algorithms and Section IV provides
illustrative results. Section V concludes the study.
(a) c1 and c2 share backup lightpath l4 (b) backup lightpaths l3 and l4 can
II. E LASTIC S EPARATE P ROTECTION AT C ONNECTION share the same wavelength

In this section, we first give a brief review of shared protec- Fig. 2. Traditional shared protection in WDM network c1 :1→2, c2 :5→6,
tion in WDM networks (non-elastic shared protection). Then both of which are full-wavelength capacity.
we introduce the spectrum assignment schemes in ESPAC and
we describe it with examples. B. Spectrum Assignment Schemes in ESPAC
The flow on a lightpath is bounded by the maximum
A. Traditional Shared protection capacity of a transponder. Correspondingly, the spectrum used
We explain shared protection in WDM networks in Fig. 2. by a lightpath, adjusted to the flow on it, is limited by a
Assume there are two connections: c1 is from node 1 to 2, spectrum range. For example, in Fig. 3, the spectrum range
and c2 is from node 5 to 6, both of which are full-wavelength is 8 slots. If we assume the spectrum used by a lightpath is
capacity. In Fig. 2(a), c1 ’s working path traverses lightpath l1 , allocated around its central frequency, then the transponder can
and its backup path is the concatenation of lightpaths l3 , l4 , be tuned so that the lightpath occupies 2/4/6/8 slots (shown
and l5 . c2 ’s working path traverses l2 , and its backup path is as different degree of gray). We define a wavelength channel
the concatenation of l6 , l4 , and l7 . Since l1 and l2 are link as the spectrum range of a lightpath with its guard bands. For
disjoint, their backup paths can share l4 . An alternative shared example, in Fig. 3, guard band is 2 slots, and a wavelength
protection is shown in Fig. 2(b): l1 and l2 are protected by l3 channel is 10 slots.
and l4 respectively. On link 3→4, l3 and l4 can share the same We use First-Fit spectrum assignment for the routing of the
wavelength because their working paths are link disjoint. The Working flows (FFW), and Last-Fit spectrum assignment to
difference between these two is the number of transponders route the Backup flows from the other end of the fiber (LFB).
used for backup provisioning: in Fig. 2(a), we need five pairs Between the wavelengths assigned to carry backup flow, we
of transponders, while in Fig. 2(b), we need two pairs. allow spectrum overlap. For example, in Fig. 1, the number
Figure 2 represents two sharing opportunities offered by of frequency slots that backup wavelength channels overlap
non-elastic shared protection. (1) If the working paths of two with each other is 3. The overlapped frequency slot is shared
connections are link disjoint, they can share the same backup between adjacent wavelength channels in the time domain: it is
lightpath (Fig. 2(a)). (2) If the working paths of two backup occupied by at most one of the adjacent wavelengths at a time.
lightpaths are link disjoint and these two backup lightpaths If the flow on lightpath 2 decreases but the flow on lightpath
go through the same physical link, then they can share the 1 increases, the elastic transponder can be tuned as shown in
same wavelength on that link (Fig. 2(b)). But since in WDM the callout in Fig. 1, where the shaded region represents the
networks, each wavelength channel is limited within its rigid used spectrum. Enough guard band must be satisfied between
grid, non-elastic shared protection cannot provide spectrum the flows on adjacent wavelengths.
sharing between adjacent wavelength channels.
C. Examples
However, enabled by the elasticity of the transponders,
ESPAC offers a new opportunity of spectrum sharing: if the Below, we illustrate ESPAC via examples on NSFNET
working paths of two connections are link disjoint, and if their in Fig. 4, where each edge corresponds to a bidirectional
backup paths traverse two lightpaths which are adjacent on a fiber. We make the following assumptions: each frequency
fiber link, then the two backup lightpaths can share spectrum. slot utilizes 12.5 GHz; the modulation level for used slot is 4
We introduce spectrum assignment schemes in ESPAC next. bits/symbol; when additional flow is groomed into an existing

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

(a) spectrum utilization on links 2 → 4/4 → 5 after provisioning c1

(b) spectrum utilization on links 2 → 4/4 → 5 after provisioning c2


Fig. 3. Spectrum assignment of a wavelength channel.

(c) spectrum utilization on links 2 → 4/4 → 5 after provisioning c3

(d) spectrum utilization on links 2 → 4/4 → 5 after provisioning c3 if


using traditional shared protection

Fig. 5. Spectrum utilization of provisioning c1 , c2 , and c3 .


(a) after provisioning c1
the spectrum range of l2 , and dark gray represents the backup
spectrum reserved on l2 for c1 , i.e., 2 slots.
Suppose c1 remains in the network when c2
h1, 8, 400 Gbpsi arrives, which needs 8 slots. One solution
of provisioning c2 is shown in Fig. 4(b). The working path
traverses lightpath l4 , and its backup path traverses lightpaths
hl5 , l2 , l6 i. c2 ’s working path (i.e., l4 ) is link disjoint with
c1 ’s working path (i.e., l1 ), and their backup paths are both
routed on l2 , so they can share 100-Gbps capacity on l2 ,
(b) after provisioning c2 shown in Fig. 5(b). If l1 fails, c1 will be rerouted on l2 so
that its backup flow consumes spectrum from A to B in Fig.
5(b). If l4 fails, c2 will be rerouted on l2 so that its backup
flow consumes spectrum from C to D in Fig. 5(b).
Suppose c1 and c2 remain in the network when c3
h3, 10, 400 Gbpsi arrives, which needs 8 slots. One way of
provisioning c3 is shown in Fig. 4(c). Its working path
traverses lightpath l7 and its backup path traverses lightpaths l8
and l9 . Because c3 ’s working path (i.e., l7 ) is not link disjoint
with c1 ’s working path (i.e., l1 ), their backup lightpaths l2
(c) after provisioning c3 and l8 cannot be assigned on the same wavelength. Under
ESPAC, we can assign them on adjacent wavelength channels
Fig. 4. ESPAC: provisioning connection c1 (h2, 6, 100 Gbpsi), overlapped with each other shown in Fig. 5(c). If link 3→6
c2 (h1, 8, 400 Gbpsi), and c3 (h3, 10, 400 Gbpsi).
fails, c1 will be rerouted on l2 so that its backup flow consumes
spectrum from A to B; c3 will be rerouted on l8 such that its
lightpath, the spectrum is allocated around the existing flow backup flow consume spectrum from E to F . So even if link
centered by the central frequency; the maximum capacity of 3→6 fails, there is still a 2-slot guard band between l2 and
elastic transponder is 400 Gbps, which needs 8 frequency slots; l8 . Note that the working path of c2 (i.e., l4 ) is link disjoint
guard band is 2 slots; and the overlapped spectrum shared by with the working path of c3 (i.e., l7 ), so their backup paths
adjacent backup wavelength channels is 3 slots, which means can share the spectrum overlap: if l4 fails, c2 will be rerouted
the interval between adjacent central frequencies is 8+2-3=7 on l2 so that its backup flow takes spectrum from C to D;
slots. A connection request c is represented as hs, d, Bi, which if l7 fails, c3 will be rerouted on l8 such that its backup flow
specifies the source node, destination node, and bandwidth takes spectrum from E to F . But since l4 and l7 will not fail
requirement in terms of Gbps. at the same time (single failure assumption), their wavelength
When the first connection request c1 h2, 6, 100 Gbpsi ar- channels can overlap 3 frequency slots with each other on
rives, 2 frequency slots are needed to carry it. One way spectrum and it is guaranteed that the spectrum overlap is
of provisioning c1 is shown in Fig. 4(a). Its working path used by only one of them at one time.
traverses lightpath l1 , the solid line, and its backup path Now we show in Fig. 5(d) the spectrum utilization on link
traverses hl2 , l3 i, the dashed lines. l2 utilizes spectrum on links 2 → 4/4 → 5 after provisioning c3 if we use traditional
2→4 and 4→5, shown in Fig. 5(a), where dashes represent shared protection under the assumption that we use the same

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

routing as ESPAC. Because c3 ’s working path (i.e., l7 ) is not


link disjoint with c1 ’s working path (i.e., l1 ), their backup
lightpaths l2 and l8 cannot be assigned on the same wave-
length. If we assign l2 and l8 as adjacent wavelength channels,
the wavelength channels must be side by side and separated
by the 2-slot guard band, because using traditional shared
protection, wavelength channels cannot share spectrum with
each other. Therefore, using ESPAC, we save three frequency
slots, compared with traditional shared protection.
In the previous example, we have shown how ESPAC can
enable backup sharing both as in traditional shared protection
and also among spectrum adjacent connections. ESPAC is
enabled by the expansion and contraction of transponders in
elastic networks, which allows us to break the fixed boundary
between lightpaths on spectrum and explore more flexible
utilization of bandwidth resources.
III. H EURISTIC A LGORITHMS Fig. 6. Graph representation of node 3 in Fig. 4(c).
In this section, we propose a heuristic approach to solve TABLE I
ESPAC under dynamic traffic. C OST A SSIGNMENT S CHEME IN A LG . 1
edge e κ(e)
A. Grooming-Node Model and Network-State Representation layerIndex < W · p 10−5 *layerIndex+1
wavelength
layerIndex > W · p 10−5 *(W + 1−layerIndex)+1
We adopt an auxiliary graph [17] to solve the routing and lightpath 0.7*hopNumber
spectrum assignment problem in ESPAC. We use the terms transponder 0.01
vertex and edge in the auxiliary graph, corresponding to node
and link in the original network. A node in the original network
relates to W + 1 vertices λ0 , λ1 , · · · , λW in the auxiliary assignment. So, as input, we are also given a parameter p,
graph. λ0 is on lightpath layer, and λi is on wavelength i.e., the proportion of wavelengths used for working flow. In
layer i (i = 1, 2, · · · , W here and in the following) (W other words, among all W wavelengths, we use the first W · p
is the number of wavelengths carried by each fiber). Each wavelengths to route the working flow, and the last W ·(1−p)
wavelength in our hypothesis of spectrum utilization identifies wavelengths to route the backup flow. OVLP is the number of
a central frequency. There are three kinds of directed edges overlapped slots between backup wavelength channels. Fig. 1
in the auxiliary graph. 1) Transponder edge: hλ0 , λi i as the shows an example of FFW-LFB where GB=2 slots, p = 0.5,
transmitter, and hλi , λ0 i as the receiver. 2) Wavelength edge: and OVLP is 3 slots. The left three wavelengths are used as
a link in the original graph is represented as W wavelength working, and the interval between adjacent central frequencies
edges from vertex λi at its source node to vertex λi at its is 10 slots. The right three wavelengths are used as backup,
destination node at each wavelength layer. 3) Lightpath edge: a and the interval between adjacent central frequencies is 7 slots.
lightpath is represented as an edge from vertex λ0 at its source Every edge e in the auxiliary graph is associated with
node to vertex λ0 at its destination node. As an example, the two attributes: available capacity and cost, which is denoted
state of node 3 in Fig. 4(c) is illustrated in Fig. 6. Assume as κ(e). Table I shows the cost assignment scheme in our
l1 takes wavelength 1, l7 takes wavelength 2, and l8 occupies study. We use FF to route working flow, so the cost of the
wavelength 9. In Fig. 6, the capacities of the four gray edges wavelengths from layer 1 to W · p increase as layer index
are zero, indicating the corresponding wavelengths are no increases. The factor 10−5 before the layerIndex is a small
longer free. value so that the paths with high layer index still have chances
By modeling every grooming node as above, the current net- to be chosen. Similarly, we use LF to route backup flow,
work state, including wavelength usage and existing lightpath thus the cost of the wavelengths from layer W · p to W
information, can be represented as one auxiliary graph. decrease as the layer index increases. The cost of a lightpath
is proportional to its hop number, so that, in Step 2 in Alg.
B. Route Computation 1, the algorithm will not choose a working path, routed on
A formal representation of the approach for route compu- existing lightpaths but going through many hops, failing to
tation is shown in Alg. 1. We are given the physical topology find a backup path. The cost of a transponder is 0.01, which is
represented by a directed graph G = (V, E, W ), where V more dominant than 10−5 , the factor before layerIndex, so that
denotes the set of nodes, equipped with BV-WXCs, E denotes the algorithm will first choose a list of spectrum-continuous
the set of directed single fibers. Initially, in Step 1, there are paths, and then apply FF for working and LF for backup.
|V |·(W +1) vertices, 2W ·|V | transponder edges, and W ·|E| To compute the K minimal-cost paths for the working path
wavelength edges in Ga . In response to each connection in Step 2, we do not consider the existing backup lightpaths
request, lightpath edges may be added. Guard band is given as or wavelengths with layer index larger than W · p because
GB. As discussed in Section II, we use FFW-LFB spectrum 1) under ESPAC, a lightpath carries either working flow or

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

Algorithm 1 ESPAC
Require: G = (V, E, W ), c = hs, d, B, ti, routing and
spectrum usage of existing lightpaths/connections, GB, C,
OVLP, and K.
Ensure: link-disjoint working and backup paths, return
NULL if no such paths are found.
1: construct the auxiliary graph Ga to represent the current
Fig. 7. An example of Cmax (l1 ) < C.
network state (including wavelength usage and existing
lightpath information), as shown in Section IV.A.
2: compute K minimum-cost paths Pw = {pkw |1 ≤ k ≤ lightpaths or wavelengths with layer index smaller than W · p.
K 0 , 0 ≤ K 0 ≤ K} in Ga from the lightpath-layer vertex The free capacity of a backup wavelength/lightpath edge e
of node s to the lightpath-layer vertex of node d based on in this step is denoted as Cb (e), where Cb (e) = Cmax (e) −
the following edge-cost function κw (e) (if there are less Cbackup (e), Cmax (e) is the maximum capacity that can be
than K paths between the vertices, then the algorithm will carried on e, and Cbackup (e) is the backup capacity already
compute all the K 0 eligible paths, 0 ≤ K 0 ≤ K); return reserved on e (Cbackup (e) = 0 if e is a wavelength). Here,
NULL if Pw is empty, Cmax (e) may not be as much as C, an example of which
1) if e represents a transponder, then κw (e) = κ(e) is shown in Fig. 7. Lightpath l1 has an adjacent lightpath l2 .
2) otherwise Since l2 is using the overlap between l1 and l2 , the maximum
(a) if Cw (e), the free capacity of e is less than capacity of l1 is s2 , which is smaller than C, marked as s1 .
B or the wavelength layer index of e is larger Thus, to compute Cmax (e) of a backup wavelength/lightpath
than W · p, κw (e) = ∞ e, we need to first compute Cbackup (l), where l is the adjacent
(b) otherwise, κw (e) = κ(e) backup lightpath of e, and Cbackup (l) is the total backup
capacity that has been reserved for backup flow. To compute
3: k = 1,
Cbackup (l), we associate l with a conflict set νl represented
4: while k ≤ K 0 and backup path pkb is not found do 0 0 0
as an integer set, {νle |∀e0 ∈ E, 0 ≤ νle ≤ C}, where νle
5: for candidate working path pkw , compute a minimum-
represents the amount of flow that will be rerouted on lightpath
cost link-disjoint path pkb from the lightpath-layer vertex
l when link e0 fails:
of node s to the lightpath-layer vertex of node d based
0
on the following edge-cost function κb (e): Cbackup (l) = max∀e0 ∈pkw {νle }, (1)
1) if e does not represent a lightpath or a wavelength,
where pkw is the kth candidate working path calculated in
then κb (e) = κ(e)
Step 2. In traditional shared protection, Cmax (e) is C when
2) otherwise
e denotes a backup wavelength/lightpath, and the spectrum
(a) if Cb (e), the free capacity of e is less than B reserved on one backup lightpath does not affect the maximum
or the wavelength layer index of e is smaller capacity of its adjacent wavelengths and lightpaths.
than W · p, or e is not link disjoint with pkw
κb (e) = ∞ IV. I LLUSTRATIVE N UMERICAL R ESULTS
(b) otherwise, κb (e) = κ(e) A. Network and Traffic Model
6: k=k+1 if pkb is not found,
We simulate a dynamic network environment with the
7: end while
assumptions that the connection arrival process is Poisson and
8: return NULL if pkb is not found,
the connection holding time follows a negative exponential
9: update network state: update the spectrum usage of light-
distribution with unit (normalized) mean. Connections requests
paths involved in the paths pkw and pkb ; set up new
are unidirectional and uniformly distributed among all node
lightpaths (if necessary).
pairs. 100,000 connections are simulated in each experiment.
10: for every lightpath l that pkb traverses, νle ← νle + B for
We consider a typical backbone network topology in Fig. 4.
every link e used by the lightpaths that pkw traverses,
Assume each fiber has a total spectrum of 4400 GHz (the same
11: return pkw as the working path and pkb as the backup path.
as 88 wavelengths with 50 GHz grid as in WDM networks).
The capacity of the elastic transponder is 400 Gbps. The
spectrum range of a lightpath is 8 slots. We assume the guard
backup flow, but not both; and 2) all wavelengths with layer band is 25 GHz, which is 2 frequency slots. To quantitatively
index larger than W · p are reserved for backup flow routing present the performance of our approach, we use bandwidth-
in our approach. The maximum capacity of the transponder is blocking ratio (BBR), defined as the amount of bandwidth
C. The free capacity of a working wavelength/lightpath edge blocked over the amount of bandwidth offered.
e in this step is denoted as Cw (e), where Cw (e) = C if e is a In our simulations, we consider the effect of two parameters:
wavelength. Cw (e) = C − Cworking (e) if e is a lightpath and p and OVLP. Table II gives the total number of wavelengths
Cworking (e) is the current working flow on e. and total spectrum reserved as backup on a fiber link according
When we compute the minimal-cost path for the backup to different p and OVLP values. The bigger p and OVLP,
path in Step 5, we do not consider all the existing working the less total backup spectrum we reserve. Here we want to

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

TABLE II
N ETWORK R ESOURCE A SSIGNMENT UNDER D IFFERENT S IMULATION
S ETTINGS
Total number Total backup spectrum
OVLP of wavelengths (in terms of GHz)
p = 0.5 p = 0.6 p = 0.5 p = 0.6
0 35 35 2175 1750
1 37 36 2037.5 1587.5
2 39 38 1925 1525
3 41 39 1787.5 1350
4 43 41 1625 1250

emphasize that, when OVLP=0, it means adjacent wavelength


channels do not share spectrum with each other, which,
in terms of spectrum utilization, is the same as traditional
shared protection in a 400-Gbps WDM network where each (a) p = 0.5
wavelength takes 125 GHz (100 GHz is spectrum range and
25 GHz is guard band). Though 400-Gbps WDM network
is a challenge for the technology today, here as we focus
on spectrum utilization, we consider ESPAC OVLP=0 to be
the same with 400 Gbps WDM. So when we compare BBR
with OVLP varying from 0 to 4, we are essentially comparing
shared protection in WDM network (OVLP=0) with ESPAC
under different spectrum sharing between adjacent lightpaths
(OVLP=1,2,3,4). With this in mind, let us look into the results.

B. Results
We first present a set of results where all the connections
request are 100 Gbps. Fig. 8(a) reports the results for p = 0.5,
which means the number of working wavelengths is the
same as that of backup wavelengths. The scenario “100 Gbps (b) p = 0.6
WDM” represents traditional shared protection in a 100-Gbps
WDM network, where a connection will fully fill in one
wavelength. For other scenarios with “OVLP” values, each
lightpath is 400 Gbps. We can see that because grooming
100-Gbps connection unto 400-Gbps lightpath uses much less
spectrum as guard band, results with OVLP values have much
lower BBR than 100 Gbps WDM. Among the ESPAC results,
the best OVLP value depends on the network offered load.
When the traffic load 1100 Erlang, the larger the OVLP, the
more efficiently the spectrum is utilized (i.e., OVLP=4, where
the reserved backup spectrum is 1625 GHz according to Table
II). When the traffic load ≤1100 Erlang, a “medium”-size
overlap spectrum gives the lowest BBR (i.e., OVLP=2 or 3,
where the reserved backup spectrum is 1925 GHz or 1787.5
GHz according to Table II). (c) FFW-LFB with different p values and FF with different
overlap values
The BBR for p = 0.6 are given in Fig. 8(b). We observe
different rules of best OVLP values from p = 0.5. When the Fig. 8. BBR versus network offered load for connection = 100 Gbps.
traffic load ≥1200 Erlang, 2-slot backup sharing gives the
lowest BBR. When load ≤1100 Erlang, 1-slot backup sharing
is the most beneficial. overlap backup spectrum, the BBR is higher because there
In Fig. 8(c), a point of the three FFW-LFB scenarios is not enough spectrum to route working flow. If we reserve
corresponds to the lowest BBR of a specific p value, with less backup spectrum, which means larger p value or more
its corresponding OVLP value shown beside the point. We overlap backup spectrum, the BBR is higher because there
observe that the network setting of p = 0.6 always has the is not enough spectrum to route backup flow. In Fig. 8(c),
lowest BBR with OVLP=1 or 2. According to Table II, in we also compare the FFW-LFB spectrum assignment with FF.
such scenarios, the reserved backup spectrums are 1587.5 GHz Under FF, working and backup traffics are assigned from the
and 1525 GHz out of 4400 GHz on a fiber. If we reserve same end of the fiber without differentiation. For all traffic
more backup spectrum, which means smaller p value or less loads, FFW-LFB p = 0.6 has the lowest BBR. When the

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.

traffic load ≤1200 Erlang, FF OVLP=0 has higher BBR but


outperforms FFW-LFB p=0.5 and p=0.7. When the traffic
load ≥1300 Erlang, all FFW-LFB scenarios outperform FF.
Thus, FFW-LFB with appropriate backup overlap is best fit
for elastic network especially when the network load is high.
We also show how different overlap values between adjacent
wavelengths effect BBR under FF, and observe that allowing
overlap under FF is not beneficial, i.e., FF OVLP=0 gives the
lowest BBR among all FF scenarios.
In our results shown in Fig. 9, the number of connection
requests follows the distribution 100 Gbps : 200 Gbps : 300
Gbps : 400 Gbps = 8 : 4 : 2 : 1. Fig. 9(a) reports the results
for p =0.6. For all traffic loads, 1-slot overlap gives the lowest
BBR. When traffic load is low (i.e., 450 Erlang), OVLP=0 has
(a) p = 0.6
closest BBR with OVLP=1. When traffic load is high (i.e., 650
Erlang), OVLP=2 has closest BBR with OVLP=1.
In Fig. 9(b), a point of the three FFW-LFB scenarios
corresponds to the lowest BBR of a specific p value under
a specific network load using ESPAC, with its corresponding
OVLP values shown beside the point. Among these three
scenarios, p =0.6 always has the lowest BBR with OVLP=1.
For p=0.7, OVLP=0 always gives the lowest BBR, and when
traffic load is high (i.e., 650 Erlang), it has close BBR with
OVLP=1 p=0.6. We also compare FFW-LFB with FF in this
setting: FFW-LFB p=0.6 outperforms all FF scenarios. When
traffic load ≤550 Erlang, the BBR of FF OVLP=0 is higher
than FFW-LFB p=0.6 but is lower than the other two FFW-
LFB scenarios. When traffic load ≥600 Erlang, FFW-LFB
outperforms all FF scenarios. Thus, in this heterogeneous
(b) FFW-LFB with different p values and FF with different
traffic profile, FFW-LFB with appropriate backup overlap is overlap values
also best fit for elastic network.
Fig. 9. BBR versus network offered load for heterogeneous connection
V. C ONCLUSION requests.

We investigated the survivable traffic grooming problem


for elastic optical networks. Based on a new elastic-width [6] B. Kozicki, H. Takara, K. Yonenaga, and M. Jinno, “Efficient elastic
spectrum assignment scheme, enabled by the elasticity of the optical path network for transmission beyond 100G,” Proc., SPIE, 2011.
[7] L. Velasco, et al., “Elastic spectrum allocation for variable traffic in
optical transponder, we proposed elastic separate protection at flexible-grid optical networks,” Proc., OFC/NFOEC 2012.
connection (ESPAC), exploring sharing spectrum between ad- [8] A. Eira, J. Pedro, and J. Pires, “Optimized design of shared restoration
jacent backup lightpaths – for grooming a connection request in flexible-grid transparent optical networks,” Proc., OFC/NFOEC 2012.
[9] X. Shao, et al., “Shared-path protection in OFDM-based optical net-
with shared protection against single fiber failures. We also works with elastic bandwidth allocation,” Proc., OFC/NFOEC 2012.
proposed a First-Fit-for-working, Last-Fit-for-backup spec- [10] A. N. Patel, et al., “Survivable transparent flexible optical WDM
trum assignment, specifically to ESPAC. Our results show that (FWDM) networks,” Proc., OFC/NFOEC 2011.
[11] Y. Sone, et al., “Bandwidth squeezed restoration in spectrum-sliced
this new backup sharing in elastic network context achieves elastic optical path networks (SLICE),” IEEE/OSA Journal of Optical
significant gain on spectrum saving. Communications and Networking, vol. 3, pp. 223-233, 2011.
[12] K. Christodoulopoulos, I. Tomkos, and E. A. Varvarigos, “Elastic Band-
ACKNOWLEDGMENT width Allocation in Flexible OFDM-Based Optical Networks,” Journal
of Lightwave Technology, vol. 29, no. 9, pp. 1354-1366, 2011.
We thank Rui Wang for his help with this study. [13] K. Zhu and B. Mukherjee, “Traffic grooming in an optical WDM mesh
network,” IEEE Journal on Selected Areas in Communications, vol. 20,
R EFERENCES no. 1, pp. 122-133, 2002.
[14] Y. Zhang, et al., “Traffic grooming in spectrum-elastic optical path
[1] M. Jinno, et al., “Spectrum-efficient and scalable elastic optical path networks,” Proc., OFC/NFOEC 2011.
network: architecture, benefits, and enabling technologies,” IEEE Com- [15] G. Zhang, M. D. Leenheer, and B. Mukherjee, “Optical grooming in
munications Magazine, vol. 47, pp. 66-73, 2009. OFDM-based elastic optical networks,” Proc., OFC/NFOEC 2012.
[2] K. Christodoulopoulos, et al., “Value analysis methodology for flexible [16] C. Ou, et al., “Traffic grooming for survivable WDM networks-shared
optical networks,” Proc., ECOC 2011. protection,” IEEE Journal on Selected Areas in Communications (JSAC),
[3] S. Zhang and B. Mukherjee, “Energy efficient dynamic provisioning for vol. 21, pp. 1367-1383, 2003.
spectrum elastic optical networks,” Proc., ICC 2012. [17] H. Zhu, H. Zang, K. Zhu, and B. Mukherjee, “A novel generic graph
[4] P. Wright, A. Lord, and S. Nicholas, “Comparison of optical spectrum model for traffic grooming in heterogeneous WDM mesh networks,”
utilization between flexgrid and fixed grid on a real network topology,” IEEE/ACM Transactions on Networking, vol. 11, no. 2, pp. 285-299,
Proc., OFC/NFOEC 2012. 2003.
[5] M. Eiselt, et al., “Programmable modulation for high-capacity net-
works,” Proc., ECOC 2011.

Copyright (c) 2011 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

You might also like